1. Field of the Invention
The present invention relates to an image encoder/decoder, in particularly to an image encoder/decoder for encoding/decoding digital image data at a high efficiency.
2. Description of the Related Art
In image encoding, a method for making the image quality of a specific area better than that of the other areas is hitherto proposed.
In the case of the method disclosed in the Japanese Patent Application Laying Open (KOKAI) No. 5-130603, for example, the minimum quantization step size is used in the area of a speaker among a plurality of persons in a videoconference and the quantization step size is controlled in an area other than the area of the speaker in accordance with a bit rate. Similarly, in the case of the method described in the document "A Study on Picture Quality Improvement Technique for Video Codec" (SHARP Technical Report, No. 6, December 1994, pp. 25-30), the image quality of the facial region is made better than that of other region by detecting the facial region out of an image and decreasing the quantization step size specified in macro-blocks in the detected area.
Moreover, in the case of the method disclosed in the Japanese Patent Application Laying Open (KOKAI) No. 5-75867, the image quality of important areas in an image is improved by using all coefficients of DCT (discrete cosine transform) to encode a specified area and using only some of the DCT coefficients to encode the other areas.
Any method is characterized by making the image quality of a selected area better than that of the other areas without changing the standard encoding method represented by H. 261.
Moreover, it is hitherto studied to store the image of background in a memory and apply it to predictive encoding in the case of encoding for a videophone. For example, in the case of the method disclosed in the Japanese Patent Application Laying Open (KOKAI) No. 64-32788, a portion hidden behind a person in a background image is successively added to a background-image memory whenever the portion appears according to movement of the person. Thereby, a more perfect background image can be obtained.
The above mentioned methods of controlling the quantization step size or limiting the number of DCT coefficients, in order to make the image quality of a selected area better than that of the other areas without changing the standard encoding method, cannot control the image quality due to difference of other encoding parameters. Specifically, the number of pixels in a frame (spatial resolution) and a sub sampling rate of a frame (time resolution) are the same in both of the selected area and the other areas. Therefore, the above image encoding method cannot make the image quality of the selected area better than that of the other areas by adjusting encoding parameters other than the quantization step size and DCT coefficients.
Moreover, the method of storing a background image in a memory assumes that the background image is stationary. Actually, however, a dynamic image having movement in the background may be encoded when a person moves behind a speaker or a videophone is used in a car, and thereby a scene outside of a window is included in the background. In this case, it is necessary to give a slight movement to the background image instead of completely stopping it.